Induced Derangement in Bone Mineral Density in Patients With

www.nature.com/scientificreports

OPEN Identifcation of a key gene module associated with glucocorticoid- induced derangement in bone mineral density in patients with asthma Suh-Young Lee1, Ha-Kyeong Won1, Byung-Keun Kim 3, Sae-Hoon Kim4, Yoon-Seok Chang4, Sang-Heon Cho1,2, H. William Kelly5, Kelan G. Tantisira6,7 & Heung-Woo Park1,2,6*

Derangement in bone mineral density (BMD) caused by glucocorticoid is well-known. The present study aimed to fnd key biological pathways associated with low BMD after glucocorticoid treatment in asthmatics using gene expression profles of peripheral blood cells. We utilized immortalized B cells (IBCs) from 32 childhood asthmatics after multiple oral glucocorticoid bursts and peripheral blood mononuclear cells (PBMCs) from 17 adult asthmatics after a long-term use of oral glucocorticoid. We searched co-expressed gene modules signifcantly related with the BMD Z score in childhood asthmatics and tested if these gene modules were preserved and signifcantly associated with the BMD Z score in adult asthmatics as well. We identifed a gene module composed of 199 genes signifcantly associated with low BMD in both childhood and adult asthmatics. The structure of this module was preserved across gene expression profles. We found that the cellular metabolic pathway was signifcantly enriched in this module. Among 18 hub genes in this module, we postulated that 2 genes, CREBBP and EP300, contributed to low BMD following a literature review. A novel biologic pathway identifed in this study highlighted a gene module and several genes as playing possible roles in the pathogenesis of glucocorticoid- induced derangement in BMD.

Derangement in bone mineral density (BMD) caused by glucocorticoids is well-known. Glucocorticoid treatment is the most common cause of secondary osteoporosis1 and can induce a dosage-dependent reduction in bone mineral accretion in childhood asthmatics2. We previously reported that polymorphisms located in CRHR1, TBCD and RAPGEF5 were associated with low BMD in childhood asthmatics afer a long-term treatment of glu- cocorticoids3–5. However, so far, there has been no comprehensive study to elucidate biologic mechanisms underlying glucocorticoid-induced derangement in BMD in asthmatics who are frequently treated by glucocorticoid. A network based analysis is an emerging and promising approach and provides new insights into BMD genetics by grouping genes into modules which are enriched for particular biological processes6,7. Te present study aimed to fnd key biological pathways associated with low BMD afer glucocorticoid treatment in patients with asthma using weighted gene co-expression network analysis (WGCNA). For this purpose, we used gene expression profles of immortalized B cells (IBCs) from childhood asthmatics afer multiple oral glucocorticoid bursts and those of peripheral blood mononuclear cells (PBMCs) from adult asthmatics afer a long-term use of oral glucocorticoids. Peripheral blood is an easily accessible tissue, although

1Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Republic of Korea. 2Institute of Allergy and Clinical Immunology, Seoul National University Medical Research Center, Seoul, Republic of Korea. 3Department of Internal Medicine, Korea University Medical Center Anam Hospital, Seoul, Republic of Korea. 4Department of Internal Medicine, Seoul National University Bundang Hospital, Seongnam, Gyeonggi-do, Republic of Korea. 5Department of Pediatrics, University of New Mexico Health Sciences Center, Albuquerque, NM, USA. 6The Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA. 7Division of Pulmonary and Critical Care Medicine, Department of Medicine, Brigham and Women’s Hospital, Boston, MA, USA. *email: [email protected]

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 1 www.nature.com/scientificreports/ www.nature.com/scientificreports

Childhood asthmatics† Adult asthmatics‡ (n = 32) (n = 17) Gender, male (%) 23 (71.9) 6 (35.3) Age, year 8.63 (1.93) 60.11 (11.71) Smoking, yes (%) Not applicable 3 (17.65) Body mass index, kg/m2 17.71 (3.21) 23.72 (5.18) Vitamin D, ng/mL 31.81 (9.97) 14.71 (7.72) Tanner stage (%) I/II/III/IV/V 24 (75.0)/6 (18.8)/1(3.1)/1 (3.1) Not applicable Baseline BMD Z-score −0.10 (2.23) Not available Ethnicity (%) Non-Hispanic white 32 (100) 0 Asian 0 17 (100) Cumulative dose of PD, mg 1188.40 (608.05) 3130.53 (1552.09) Final BMD Z-score −0.29 (0.81) −0.97 (1.37)

Table 1. Characteristics of subjects enrolled. Data indicate mean (Standard deviation); BMD, Bone mineral density; PD, Prednisone; †Characteristics measured at enrollment of the CAMP study except cumulative dose of prednisone and the fnal BMD Z-score; ‡Characteristics measured at enrollment of the present study.

Figure 1. Co-expression modules identifed in gene expression profles of Sham-treated immortalized B cells from childhood asthmatics. Top panel shows a dendrogram of 5,000 genes. Bottom panel shows colors corresponding to the cluster membership labels for each gene.

it is not directly involved in bone biology. However, it was suggested that molecular profling of peripheral blood might refect physiological and pathological events occurring in diferent tissues of the body8. Moreover, recent reports showed that PBMCs could be used to assess glucocorticoid sensitivity or efcacy of osteoporosis treat- ment9,10. Taken together, we may use gene expression profles of peripheral blood cells for the purpose of search- ing what biologic pathways are related to glucocorticoid-induced derangement in BMD. Results A total of 32 childhood asthmatics whose IBCs and BMD profles were available and 17 adult asthmatics enrolled (Table 1). Applying WGCNA to the 5,000 genes expressed in Sham-treated IBCs, we could identify 10 modules with various sizes ranging from 125 genes in the purple module to 1,372 genes in the turquoise module (Fig. 1). To emphasize the impact of strong correlations over weak ones in the network construction, we chose an empir- ical sof threshold of 6, representing a strong model ft for scale-free topology (R2 > 0.8, Figure S1). A total of 63 genes could not be assigned to a module as a membership gene, and were grouped into the grey module which was not considered for further analysis. Eigengene values of two modules (blue and magenta modules) showed signifcant associations with the BMD Z score in multivariate linear regression analysis. Eigengene value of the blue module consisted of 794 genes was negatively associated with the BMD Z score (P = 0.00294), whereas that of the magenta module containing 199 genes showed a positive association (P = 0.000758) (Fig. 2 and Figure S2). To assess the reproducibility and generalizability, we tested if our results were replicated in other gene expression profles from adult asthmatics (Sham-treated PBMCs). Replication analysis was done in two ways: preservation of module (the consistency of network module structure across gene expression profles) and association with the target phenotype (association between corresponding eigengene values and the BMD Z score). Among 10 modules identifed in the discovery cohort, three modules (turquoise, magenta and purple modules) were signifcantly preserved in the replication cohort (Fig. 3). Module preservation statistics and P values are presented

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 2 www.nature.com/scientificreports/ www.nature.com/scientificreports

Figure 2. Associations between eigengene values of the bone mineral density Z score-associated common module and the bone mineral density Z score. (A) Childhood asthmatics. (B) Adult asthmatics. Both P values were adjusted by covariates.

Figure 3. Preservation of modules identifed in gene expression profles of Sham-treated immortalized B cells from childhood asthmatics in gene expression profles of Sham-treated peripheral mononuclear cell from adult asthmatics. Te frst (top) panel shows a heatmap of pair-wise correlations among the genes comprising the turquoise, magenta, and purple modules. Te second panel shows a heatmap of the edge weights (connections) among the genes comprising the three modules. Te third panel shows the distribution of scaled weight degrees (relative connectedness) among the genes comprising the three modules. Te fourth panel shows the distribution of node contributions (correlation to module eigengene) among the genes comprising the three modules. Genes are ordered from lef to right based on their weighted degree in the discovery cohort, so as to highlight the consistency of the network properties in the replication cohort.

in Table S1. Multivariate linear regression analysis showed that eigengene value of the magenta module was significantly associated with the BMD Z score in the replication cohort (P = 0.0104) (Fig. 2). Based on these fndings, we defned the magenta module identifed in gene expression profles from childhood asthmatics as the BMD Z score-associated common module. Among 199 membership genes in the BMD Z score-associated common module, 18 genes (ARMC5, ATP2A2, CCNK, CREBBP, EP300, EP400, GTF3C1, IPO13, MTF1, NOL8, NUP188, PCF11, RFX5, SDAD1, SETD1A, SLC25A22, UBAP2L, WDR59) were taken as hub genes due to their high connectivity and association with the BMD Z score (Figure S3 and Table S2). To evaluate in vitro perturbation of modules identifed in gene expressions of Sham-treated IBCs by Dex, we measured changes of the module structure between Sham- and Dex-treated IBCs. Among the 10 modules, only the magenta module identifed in Sham-treated IBCs was not preserved in Dex-treated IBCs (a maximal

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 3 www.nature.com/scientificreports/ www.nature.com/scientificreports

Figure 4. Co-expression network of hub genes of the bone mineral density Z score-associated common module. (A) Sham-treated IBCs from childhood asthmatics. (B) Dex-treated IBCs from childhood asthmatics. For clarity, only the edges corresponding to the Pearson correlation coefcient >0.8 were shown. Te edge width is proportional to the Pearson correlation coefcient between two nodes. Te network was visualized using qgraph R package. ARM, ARMC5; ATP, ATP2A2; CCN, CCNK; CRE, CREBBP; EP3, EP300; EP4, EP400; GTF, GTF3C1; IPO, IPO13; MTF, MTF1; NOL, NOL8; NUP, NUP188; PCF, PCF11; RFX, RFX5; SDA, SDAD1; SET, SETD1A; SLC, SLC25A22; UBA, UBAP2L; WDR, WDR59.

permutation P = 0.096, Table S3). In addition, a co-expression network of the 18 hub genes showed distinct dif- ferences between them (Fig. 4). To assign biological meaning to interpretability of the BMD Z score-associated common module, we performed pathway enrichment analyses in GO biological process categories with its membership genes. To increase the specifcity in the enrichment results, we set a more stringent threshold of overlap (intersection of query and test sets ≥ 3) and excluded consideration of inferred electronic annotations (i.e., annotations created by in silico curation methods of potentially lower quality). Te most signifcantly enriched pathway was the cellular metabolic pathway (GO:0044237, FDR adjusted P value = 2.75E-7). Tis pathway consists of 10,751 genes and 150 genes out of 199 genes belong to the BMD Z score-associated common module overlapped. Enriched pathways identifed are shown in Table 2. As we could not fnd datasets to replicate our fndings, we performed a pathway

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 4 www.nature.com/scientificreports/ www.nature.com/scientificreports

analysis using the top-200 single nucleotide polymorphisms (SNPs) (Table S4) that were signifcantly associated with the bone mineral accretion in childhood asthmatics treated by oral prednisone bursts in our previous genome-wide association study4. We performed a pathway analysis using genes mapped to 200 SNPs and identify the cellular macromolecular metabolic process (GO:0044260) as the most signifcant candidate causal pathway (FDR P value = 0.02). Discussion Te purpose of this study was to fnd key biological pathways associated with glucocorticoid-induced BMD derangement including a reduction in bone mineral accretion in childhood asthmatics and a reduction in BMD in adult asthmatics afer long-term treatment of glucocorticoid. Using gene expression profles on blood cells, we could identify a key gene module composed of 199 genes whose eigengene value showed a signifcant association with the BMD Z score in both childhood and adult asthmatics and whose connectivity was preserved as well. Membership genes in this module were significantly overrepresented in the cellular metabolic pathway (GO:0044237). A previous report exploring diferential gene expression profles on PBMC by the graph clustering approach and GO term enrichment analysis showed 9 gene clusters associated with osteoporosis11. Nine clusters included various GO pathways, such as, response to virus (GO:0009615), immune response (GO:0006955), response to hypoxia (GO:0001666), and response to extracellular stimulus (GO:0009991) which were similar with GO pathways identifed in this study. Te cellular metabolic pathway representing the chemical reactions and pathways by which individual cells transform chemical substances never been recognized in previous studies focused on osteoporosis. To get a further insight, we performed a pathway analysis using the genes mapped to top-200 SNPs that were signifcantly associated with the bone mineral accretion in childhood asthmatics in our previous genome-wide association study4. We identifed the cellular macromolecular metabolic process (GO:0044260) as the most important candidate causal pathway. Tis pathway is a sub-pathway of the cellular metabolic pathway (GO:0044237)12. Tese SNPs could not be directly linked to gene expression/modules identifed in the present study, as genomics data were generated from diferent populations. However, we thought that this result would be an additional evidence suggesting a possible role of the cellular metabolic pathway in osteoporosis pathogenesis. Cellular metabolism is an important part of the bone biology and homeostasis13,14, but, so far, no direct evidence connecting this pathway to glucocorticoid-induced BMD derangement has been issued, although the cellular amine metabolic pathway (GO:0033240), a sub-pathway of the cellular metabolic pathway, was reported to be enriched in gene expressions of PBMC from patients with osteoporosis compared to normal control15. Further studies are needed to confrm our observations. It was not easy to give biological meanings to our fndings at the pathway level, as identifed pathways were diverse and contained many genes (for example, the cellular metabolic pathway included 9416 genes12). For this reason, we selected highly connected intramodular18 hub genes. By reviewing the literatures, we selected two genes of putative interest: EP300 encoding E1A binding protein p300 and CREBBP coding CREB binding protein (CBP). CBP and p300 are highly related transcriptional co‐activators possessing histone acetyltransferase activity and are known to participate in the activities of hundreds of diferent transcription factors involved in a variety of cell functions16. Up-regulation of miR-132-3p could inhibit osteoblast diferentiation in part by decreasing p300 expression17, and CBP could mediate osteoblast-adipocyte lineage plasticity by interaction with Maf18,19. In addition, both CBP and p300 were known to be important co-factors mediating transforming growth factor-beta and bone morphogenic protein signaling which have fundamental roles in bone homeostasis20, and to modulate parathyroid hormone regulation of osteoblast21. Interestingly, it has been also reported that these two proteins are involved in steroid receptor signaling22. Steroid receptors require coactivators for efcient activation of target gene expression and these coactivators might contribute to tissue specifc responses to steroid hormone. Likewise, CBP and p300 might function additively or antagonistically to each other depending on their relative concentrations and type of target tissue, to infuence the sensitivity of tissues to glucocorticoids23. In addition, it was reported that CREBBP mutations resulted in impaired expression of GC receptor responsive gene and impaired histone acetylation and transcriptional regulation24,25. Given that CBP and p300 are involved not only in bone homeostasis but also in glucocorticoid receptor signaling, a gene module identifed in this study which contained CREBBP and EP300 may play an important role in glucocorticoid-induced derangement in BMD. Among other genes belong to the hub genes, ARMC5 encoding armadillo repeat containing 5 and IPO13 encoding importin 13 are of interest. ARMC5 inactivation afects steroid production and cell survival in vitro and adrenal hyperplasia associated with severe Cushing syndrome26. As well known, Cushing’s syndrome results in osteoporosis, which is analogue to glucocorticoid-induced osteoporosis27. IPO13 is a steroid-inducible gene and the glucocorticoid receptor is a cargo substrate for importin 1328. IPO13 is thus possibly associated with glucocorticoid signaling, although its role in glucocorticoid-induced osteoporosis is unknown. Te strength of our study was that we searched biological mechanisms underlying glucocorticoid-induced derangement in BMD with a network based methodology for the frst time. A recent review highlighted an importance of network-based approach in genomic studies for osteoporosis29. Considering that the modularity of networks is inherent to cell biology and thus molecular interactions organized into functional unit are more relevant to explain biological phenomena30, the present study might provide a comprehensive and holistic under- standing of glucocorticoid-induced derangement in BMD. To generate biological networks, we utilized WGCNA in the present study. WGCNA, unlike the other biological networks, is a way to organize data in an unbiased manner and provides a gene module summary which can be related with clinical phenotypes29. In a systems genetics aspect, if a Bayesian structure learning algorithm is to be applied to WGNCA, the direction of the fow of molecular information would be elucidated, which enables us fgure out the causal relationship. Recently, such an attempt identifed novel pathways in patients with the late-onset Alzheimer’s disease pathology and coronary artery disease31,32. However, to our knowledge, Bayesian structure learning algorithm in WGNCA has not yet been applied in the feld of bone mineral density and steroid response. Another important thing is that we can

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 5 www.nature.com/scientificreports/ www.nature.com/scientificreports

ID Name† Depth‡ P value§ GO:1903901 Negative regulation of viral life cycle 1 0.0418 GO:0045071 Negative regulation of viral genome replication 1 0.014 GO:0006475 Internal protein amino acid acetylation 1 0.0448 GO:0018394 Peptidyl-lysine acetylation 1 0.0453 GO:0018393 Internal peptidyl-lysine acetylation 2 0.0394 GO:0018076 N-terminal peptidyl-lysine acetylation 2 0.0287 GO:0016570 Histone modifcation 1 0.0291 GO:0016573 Histone acetylation 2 0.0352 GO:0010646 Regulation of cell communication 1 0.0485 GO:0051169 Nuclear transport 1 0.041 GO:0006913 Nucleocytoplasmic transport 2 0.0374 GO:0002886 Regulation of myeloid leukocyte mediated immunity 1 0.0321 GO:0002279 Mast cell activation involved in immune response 1 0.0453 GO:0051640 Organelle localization 1 0.0335 GO:0043300 Regulation of leukocyte degranulation 1 0.0336 GO:0051656 Establishment of organelle localization 1 0.0052 GO:0033003 Regulation of mast cell activation 1 0.0283 GO:0033006 Regulation of mast cell activation involved in immune response 2 0.0153 GO:0043303 Mast cell degranulation 1 0.0453 GO:0043304 Regulation of mast cell degranulation 2 0.0153 GO:0050657 Nucleic acid transport 1 0.0335 GO:0051236 Establishment of RNA localization 1 0.0354 GO:0050658 RNA transport 2 0.0335 GO:0006405 RNA export from nucleus 3 0.0287 GO:0071166 Ribonucleoprotein complex localization 1 0.0219 GO:0071426 Ribonucleoprotein complex export from nucleus 2 0.0205 GO:0008150 Biological process 1 0.00234 GO:0071840 Cellular component organization or biogenesis 2 0.0279 GO:0044085 Cellular component biogenesis 3 0.00104 GO:0008152 Metabolic process 2 0.00000288 GO:0044238 Primary metabolic process 3 0.000013 GO:0071704 Organic substance metabolic process 3 0.0000188 GO:0009058 Biosynthetic process 3 0.000516 GO:0006807 Nitrogen compound metabolic process 3 0.0000273 GO:0009987 Cellular process 2 0.000425 GO:0044237 Cellular metabolic process 3 0.000000275 GO:0043933 Macromolecular complex subunit organization 1 0.0278 GO:0071822 Protein complex subunit organization 2 0.019 GO:0065003 Macromolecular complex assembly 1 0.0289 GO:0006461 Protein complex assembly 2 0.0278 GO:0010629 Negative regulation of gene expression 1 0.0138 GO:0000122 Negative regulation of transcription from RNA polymerase II promoter 1 0.0272 GO:0002520 Immune system development 1 0.0138 GO:0048534 Hematopoietic or lymphoid organ development 2 0.0167 GO:0030097 Hemopoiesis 3 0.0225 GO:0006325 Chromatin organization 1 0.034 GO:0031648 Protein destabilization 1 0.0356 GO:0033554 Cellular response to stress 1 0.00328 GO:0036294 Cellular response to decreased oxygen levels 1 0.0448 GO:0071456 Cellular response to hypoxia 2 0.038 GO:0043618 Regulation of transcription from RNA polymerase II promoter in response to stress 1 0.0428 GO:0061418 Regulation of transcription from RNA polymerase II promoter in response to hypoxia 2 0.0466 GO:0051649 Establishment of localization in cell 1 0.0302

Table 2. Biological pathways enriched in the bone mineral density Z score-associated common module. †Gene ontology biological pathway, ‡Only depth 1~3 were presented, §Benjamini-Hocherg FDR P value.

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 6 www.nature.com/scientificreports/ www.nature.com/scientificreports

analyze co-expressions across a set of perturbations with WGCNA. In the present study, we found that a structure of the BMD Z score-associated common module was disrupted by in vitro perturbation using gene expression profles of Dex-treated IBCs. Eigengene values of this module showed signifcant positive associations with the BMD Z score in both childhood and adult asthmatic (Fig. 2). We assume that a disruption of this protective gene module by glucocorticoid may result in BMD reduction in asthmatics treated by glucocorticoid for a long time, which provides a functional relevance of our observations. In addition, we confrmed that a module signifcantly associated with the BMD Z score preserved in diferent conditions (bone mineral accretion in childhood asthmatics and osteoporosis in adult asthmatics), which may increase the generalizability of observations in this study. Tere are some limitations in our study. First, we utilized gene expression profles of blood cells instead of the cells directly involved in bone biology. We could fnd a publicly available gene expression dataset of Sham- and Dex-treated osteoblast (GSE21727)33. Tis dataset was generated from cultured trabecular bone cells obtained from ~100 unrelated Caucasian donors treated by Sham and Dex (100 nM, 2 hour later). Using this dataset, we could fnd that the magenta module (the BMD Z score-associated common module) identifed in this study was preserved in gene expressions of Sham-treated osteoblasts (a maximal permutation P = 0.0000599; Table S5 and Figure S4). Similar to Dex-treated IBCs, this module was not preserved in gene expressions of Dex-treated osteoblasts (a maximal permutation P = 0.253; Data were not shown). Tese fndings suggested that a gene modules of this study might be important across diferent tissue types (blood cells and osteoblast), although GSE21727 was a small dataset (one subject with three replicates). Evidently, previous reports showed that tissue-specifc genes can be expressed in non-tissue-specifc manner34,35, and peripheral blood cells expressed approximately over 80% of the genes encoded by the human genome36. In addition, it was demonstrated that osteoclasts could be generated from PBMC population37,38 and PBMC could be used to assess efcacy of osteoporosis treatment10. Taken together, peripheral blood cells may be a surrogate for bone tissue and recapitulate bone cell biology in part, although further studies will be needed. Secondly, a small number of participants, diferent ethnicity and age (Non-Hispanic white childhood asthmatics vs. Asian elderly asthmatics), and diferent ways of exposure to glucocorticoid (short-term burst vs. chronic use) were considered before generalizing our fndings. Previously, age-dependent alterations in osteoblast and osteoclast activity were reported39,40. In general, short-term glucocorticoid use was associated with mild adverse efects, whereas long-term use might be associated with more serious sequel including osteoporosis41. For this reason, we provided information on the hub genes and enriched biologic pathways of genes belong to the blue module, a childhood asthmatics-specifc gene module, in the supplementary information (Figure S5, Table S6, and Table S7). Currently, there are no publicly available datasets to validate our fndings. However, a replication analysis should be done when datasets become available. In addition, an integrative analysis using other potentially relevant -omics data such as, miRNAs, lncRNAs, and methylation profles is warranted in the future. In conclusion, the potentially signifcant genes and pathways identifed in our analysis may help further elucidate of the molecular mechanisms and provide novel insights regarding glucocorticoid-induced derangement in BMD in asthmatics. Methods All experimental protocols were approved by the Institutional Review Board of the corresponding institutions. Tis study was approved by the Institutional Review Board of the Brigham and Women’s Hospital (2002-P- 00331/41) and the Seoul National University Hospital (H-1508-095-695). Informed consents were obtained from all study participants and a parent and/or legal guardian, if subjects were under 18. And all methods were carried out in accordance with relevant guidelines and regulations. Te overall study design is presented in Fig. 5.

Study populations and bone mineral density measurements. Te discovery cohort consisted of Non-Hispanic white children from the Childhood asthma management program (CAMP) trial whose IBCs were available42. In asthma exacerbations, short courses of oral prednisone were prescribed per protocol to them. Each burst consisted of 2 mg/kg per day (up to 60 mg) of prednisone for 2 days followed by 1 mg/kg per day (up to 30 mg) for 2 days. Subsequent maintenance of prednisone was allowed when the improvement was not sufcient. BMD measurements of the lumbar spine (L1-L4) were performed annualy during the study period by means of dual-energy x-ray absorptiometry [Hologic (Hologic Inc., Waltham, MA, US) or Lunar (GE Healthcare, Madison, WI, US)]. Te BMD Z score that represents a diference between the measured bone density in a subject and the average bone density in age- and sex-matched controls was calculated by using the CAMP internal ref. 2. In the present study, the fnal BMD Z score measured afer mean follow-up duration of 4.3 years was used as a target phenotype. The replication cohort was drawn from adult asthmatics who were treated at Seoul National University Hospital, Seoul, Korea. All of them received more than 15 mg of prednisone per day continuously for at least one year before enrollment for the control of their symptoms. Doses above 10 mg of prednisone per day result in signifcant bone loss43. At the enrollment, BMD was measured at the lumbar spine (L1-L4) by dual-energy X-ray absorptiometry [Lunar (GE healthcare, Madison, WI, US)]. Like childhood asthmatics, the BMD Z score was used as a target phenotype.

Gene expression arrays. As previously described44 IBCs derived from childhood asthmatics from the CAMP trial were cultured in RPMI 1640 medium (Sham-treated IBCs) and treated with dexamethasone (10−6 M) for 6 hours (Dex-treated IBCs). Dex-treated IBCs were used to measure in vitro perturbation of gene expressions by Dex. PBMCs from adult asthmatics were cultured in RPMI 1640 medium for 6 hours (Sham-treated PBMCs). Gene expression levels were measured using the Illumina HumanRef8 v2 BeadChip (Illumina, San Diego, CA, US) for childhood asthmatics and the Afymetrix GeneChip Human Gene 2.0 ST (Afymetrix, Santa Clara, CA, US) for adult asthmatics. We removed probes with bad chromosome annotation and probes in X or Y

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 7 www.nature.com/scientificreports/ www.nature.com/scientificreports

Figure 5. Overall study design.

chromosome. We then did variance stabilizing transformation and quantile normalization respectively to reduce the efects of technical noises and to make the distribution of expression level for each array closer to normal distribution.

Statistical analysis. Analysis was performed with R version 3.4.3 (www.r-project.org). We performed WGCNA on each gene expression profle with the R package “WGCNA”45. To allow comparability of data from diferent microarray platforms, we frst calculated the mean expression value of each gene in two expression profles (Sham-treated IBCs and Sham-treated PBMCs) afer collapsing probes by genes. We then compared the mean expression level of genes of them and selected top-5000 genes in common with the highest correlation. Using 5,000 genes we frst characterized unsigned correlation networks and their relationships with each other and found gene modules in gene expression profles from childhood asthmatics. A module was defned as a group of highly inter- connected genes45. We computed eigengene values of modules identifed and performed multivariate linear regression analysis adjusted by baseline age, gender, BMD Z score, vitamin D level, Tanner stage, and body mass index Z score. Eigengene values can be efective and biologically meaningful tools for studying the relationships between modules of a gene co-expression network and phenotypes46. Te BMD Z score-associated module was defned when its eigengene value was signifcantly associated with the BMD Z score, a target phenotype. We next tested if the BMD Z score-associated modules identifed from childhood asthmatics and were preserved in gene expression profles from adult asthmatics using the R package “NetRep”47. NetRep can quantify the preservation of gene co-expression modules across diferent datasets and produce unbiased p values based on a permutation approach to score module preservation without assuming data are normally distributed47. Te BMD Z score-associated common module denoted a module which was identifed in gene expression profles from childhood asthmatics and signifcantly preserved in gene expression profles from adult asthmatics. We then calculated an eigengene value (the frst principle component) of the BMD Z score-associated common module in adult asthmatics and performed multivariate linear regression analysis adjusted by age, gender, vitamin D level, body mass index, and smoking status. We then performed Gene ontology (GO) pathway overrepresentation analyses of the BMD Z score-associated common modules using “g:Profler” (database version: r1730_e88_eg35)48. g:Profler (https://biit.cs.ut.ee/gpro- fler/) provides an adjusted P value calculated in a manner that accounts for the hierarchical relationships among the tested gene sets. Finally, we searched the key driving hub genes in the BMD Z score-associated common modules, as an occurrence of certain disease is less likely associated with the abnormality of the whole genes involved in the biological pathway. Hub genes were defned by module connectivity, measured by absolute value of the Pearson’s correlation (cor.geneModuleMembership > 0.8) and clinical trait relationship, measured by absolute value of the Pearson’s correlation (cor.geneTraitSignifcance >0.2)45,49. For the pathway analysis of SNPs, we selected the top-200 SNPs (Table S) from our previous report4 and put those SNPs into ICSNPathway (http:// icsnpathway.psych.ac.cn/). Te ICSNPathway enables us to identify candidate causal SNPs and pathways from genome-wide association study by one analytical framework50. Parameters used were as follow: specify SNP, P value < 1.0E-5; linkage disequilibrium cutof, r2 = 0.8; Rule of mapping SNPs to genes, within gene; Population, CEU; Pathway gene set size, minimum = 5 and maximum = 100; FDR cutof, 0.05.

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 8 www.nature.com/scientificreports/ www.nature.com/scientificreports

Data availability Due to the regulation of the Ethics Review Board of our institute, we could not deposit our data in publicly available repositories. However, immediately following publication, individual participant data that underlie the results reported in this article will be able to be shared afer de-identifcation with researchers who will provide a methodologically sound proposal. Proposals should be directed to [email protected].

Received: 21 May 2019; Accepted: 16 December 2019; Published: xx xx xxxx References 1. Weinstein, R. S. Glucocorticoid-induced bone disease. N. Engl. J. Med. 365, 62–70, https://doi.org/10.1056/NEJMcp1012926 (2011). 2. Kelly, H. W. et al. Efect of long-term corticosteroid use on bone mineral density in children: a prospective longitudinal assessment in the childhood Asthma Management Program (CAMP) study. Pediatrics 122, e53–61, https://doi.org/10.1542/peds.2007-3381 (2008). 3. Jones, T. S. et al. CRHR1 polymorphisms predict bone density in survivors of acute lymphoblastic leukemia. J. Clin. Oncol. 26, 3031–3037, https://doi.org/10.1200/JCO.2007.14.6399 (2008). 4. Park, H. W. et al. Genetic risk factors for decreased bone mineral accretion in children with asthma receiving multiple oral corticosteroid bursts. J. Allergy Clin. Immunol. 136, 1240–1246, https://doi.org/10.1016/j.jaci.2015.04.014 (2015). 5. Park, H. W. et al. A genetic factor associated with low fnal bone mineral density in children afer a long-term glucocorticoids treatment. Pharmacogenomics J. 17, 180–185, https://doi.org/10.1038/tpj.2015.92 (2017). 6. Farber, C. R. Identifcation of a gene module associated with BMD through the integration of network analysis and genome-wide association data. J. Bone Miner. Res. 25, 2359–2367, https://doi.org/10.1002/jbmr.138 (2010). 7. Chen, Y. C. et al. Integrative analysis of genomics and transcriptome data to identify potential functional genes of BMDs in females. J Bone Miner. Res. 31, 1041–1049, https://doi.org/10.1002/jbmr.2781 (2016). 8. Mohr, S. & Liew, C. C. Te peripheral-blood transcriptome: new insights into disease and risk assessment. Trends Mol. Med. 13, 422–432, https://doi.org/10.1016/j.molmed.2007.08.003 (2007). 9. Chapin, W. J. et al. Peripheral blood IRF1 expression as a marker for glucocorticoid sensitivity. Pharmacogenet. Genomics 25, 126–133, https://doi.org/10.1097/FPC.0000000000000116 (2015). 10. Stuss, M. et al. Assessment of OPG/RANK/RANKL gene expression levels in peripheral blood mononuclear cells (PBMC) afer treatment with strontium ranelate and ibandronate in patients with postmenopausal osteoporosis. J. Clin. Endocrinol. Metab. 98, E1007–1011, https://doi.org/10.1210/jc.2012-3885 (2013). 11. Xiao, H., Shan, L., Zhu, H. & Xue, F. Detection of signifcant pathways in osteoporosis based on graph clustering. Mol. Med. Rep. 6, 1325–1332, https://doi.org/10.3892/mmr.2012.1082 (2012). 12. Gene Ontology Browser, GO:0044237. Available at: http://www.informatics.jax.org/vocab/gene_ontology/GO:0044237. Accessed November 14, 2019. 13. Florencio-Silva, R., Sasso, G. R., Sasso-Cerri, E., Simões, M. J. & Cerri, P. S. Biology of Bone Tissue: Structure, Function, and Factors Tat Infuence Bone Cells. Biomed Res. Int. 2015, 421746, https://doi.org/10.1155/2015/421746 (2015). 14. Marks, S. C. Jr & Popof, S. N. Bone Cell Biology: Te Regulation of Development, Structure, and Function in the Skeleton. Am. J. Anat. 183, 1–44, https://doi.org/10.1002/aja.1001830102 (1988). 15. Li, J. J., Wang, B. Q., Fei, Q., Yang, Y. & Li, D. Identifcation of candidate genes in osteoporosis by integrated microarray analysis. Bone Joint Res. 5, 594–601, https://doi.org/10.3892/mmr.2012.1082 (2016). 16. Goodman, R. H. & Smolik, S. CBP/p300 in cell growth, transformation, and development. Genes. Dev. 14, 1553–1577 (2000). 17. Hu, Z. et al. miRNA-132-3p inhibits osteoblast diferentiation by targeting Ep300 in simulated microgravity. Sci. Rep. 5, 18655, https://doi.org/10.1038/srep18655 (2015). 18. Berendsen, A. D. & Olsen, B. R. Osteoblast-adipocyte lineage plasticity in tissue development, maintenance and pathology. Cell Mol. Life Sci. 71, 493–497, https://doi.org/10.1007/s00018-013-1440-z (2014). 19. Nishikawa, K. et al. Maf promotes osteoblast diferentiation in mice by mediating the age-related switch in mesenchymal cell diferentiation. J. Clin. Invest. 120, 3455–3465, https://doi.org/10.1172/JCI42528 (2010). 20. Wu, M., Chen, G. & Li, Y. P. TGF-β and BMP signaling in osteoblast, skeletal development, and bone formation, homeostasis and disease. Bone Res. 4, 16009, https://doi.org/10.1038/boneres.2016.9 (2016). 21. Yang, D., Guo, J., Divieti, P., Shioda, T. & Bringhurst, F. R. CBP/p300-interacting protein CITED1 modulates parathyroid hormone regulation of osteoblastic diferentiation. Endocrinology 149, 1728–1735, https://doi.org/10.1210/en.2007-0826 (2008). 22. Knutti, D., Kaul, A. & Kralli, A. A tissue-specifc coactivator of steroid receptors, identifed in a functional genetic screen. Mol. Cell Biol. 20, 2411–2422, https://doi.org/10.1128/mcb.20.7.2411-2422.2000 (2000). 23. Kino, T., Nordeen, S. K. & Chrousos, G. P. Conditional modulation of glucocorticoid receptor activities by CREB-binding protein (CBP) and p300. J. Steroid Biochem. Mol. Biol. 70, 15–25 (1999). 24. Mullighan, C. G. et al. CREBBP mutations in relapsed acute lymphoblastic leukaemia. Nature 471, 235–239, https://doi.org/10.1038/ nature09727 (2011). 25. Lu, N. Z. & Cidlowski, J. A. Glucocorticoid receptor isoforms generate transcription specifcity. Trends Cell Biol. 16, 301–307, https:// doi.org/10.1016/j.tcb.2006.04.005 (2006). 26. Assié, G. et al. ARMC5 mutations in macronodular adrenal hyperplasia with Cushing’s syndrome. N. Engl. J. Med. 369, 2105–2014, https://doi.org/10.1056/NEJMoa1304603 (2013). 27. Kaltsas, G. & Makras, P. Skeletal diseases in Cushing’s syndrome: osteoporosis versus arthropathy. Neuroendocrinology 92(Suppl 1), 60–64, https://doi.org/10.1159/000314298 (2010). 28. Tao, T., Lan, J., Lukacs, G. L., Haché, R. J. & Kaplan, F. Importin 13 regulates nuclear import of the glucocorticoid receptor in airway epithelial cells. Am. J. Respir. Cell Mol. Biol. 35, 668–680, https://doi.org/10.1165/rcmb.2006-0073OC (2006). 29. Al-Barghouthi, B. M. & Farber, C. R. Dissecting the genetics of osteoporosis using systems approaches. Trends Genet. 35, 55–67, https://doi.org/10.1016/j.tig.2018.10.004 (2019). 30. Hartwell, L. H., Hopfeld, J. J., Leibler, S. & Murray, A. W. From molecular to modular cell biology. Nature 402, C47–52, https://doi. org/10.1038/35011540 (1999). 31. Zhang, B. et al. Integrated systems approach identifes genetic nodes and networks in late-onset Alzheimer’s disease. Cell 153, 707–720, https://doi.org/10.1016/j.cell.2013.03.030 (2013). 32. Mäkinen, V. P. et al. Integrative genomics reveals novel molecular pathways and gene networks for coronary artery disease. PLoS Genet. 10, e1004502, https://doi.org/10.1371/journal.pgen.1004502 (2014). 33. Gene Expression Omnibus, GSE21727. Available at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE21727. Accessed November 14, 2019. 34. helly, J., Kaplan, J. C., Maire, P., Gautron, S. & Kahn, A. Transcription of the dystrophin gene in human muscle and non-muscle tissue. Nature 333, 858–860, https://doi.org/10.1038/333858a0 (1988).

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 9 www.nature.com/scientificreports/ www.nature.com/scientificreports

35. Kimoto, Y. A single human cell expresses all messenger ribonucleic acids: the arrow of time in a cell. Mol. Gen. Genet. 258, 233–239 (1998). 36. Liew, C. C., Ma, J., Tang, H. C., Zheng, R. & Dempsey, A. A. Te peripheral blood transcriptome dynamically refects system wide biology: a potential diagnostic tool. J. Lab. Clin. Med. 147, 126–132, https://doi.org/10.1016/j.lab.2005.10.005 (2006). 37. Komano, Y., Nanki, T., Hayashida, K., Taniguchi, K. & Miyasaka, N. Identifcation of a human peripheral blood monocyte subset that diferentiates into osteoclasts. Arthritis Res. Ter. 8, R152, https://doi.org/10.1186/ar2046 (2006). 38. Huang, H., Jiang, H., Zhang, H. & Hou, J. Generating osteoclasts from human peripheral blood mononuclear cells. Int. J. Pathol. Clin. Med. 28, 10–13, https://doi.org/10.1007/978-1-61779-415-5_11 (2008). 39. Becerikli, M. et al. Age-dependent alterations in osteoblast and osteoclast activity in human cancellous bone. J. Cell. Mol. Med. 21, 2773–2781, https://doi.org/10.1111/jcmm.13192 (2017). 40. Lai, P. et al. Loss of Rictor with aging in osteoblasts promotes age-related bone loss. Cell Death Dis. 13, e2408, https://doi. org/10.1038/cddis.2016.249 (2016). 41. Buchman, A. L. Side effects of corticosteroid therapy. J. Clin. Gastroenterol. 33, 289–294, https://doi.org/10.1097/00004836- 200110000-00006 (2001). 42. Childhood Asthma Management Program Research Group. Long-term efects of budesonide or nedocromil in children with asthma. Te Childhood Asthma Management Program Research Group. N. Engl. J. Med. 343, 1054–1063, https://doi.org/10.1056/ NEJM200010123431501 (2000). 43. Sambrook, P. N. How to prevent steroid induced osteoporosis. Ann. Rheum. Dis. 64, 176–178, https://doi.org/10.1136/ ard.2003.018556 (2005). 44. Qiu, W. et al. Pharmacogenomics: novel loci identifcation via integrating gene diferential analysis and eQTL analysis. Hum. Mol. Genet. 23, 5017–5024, https://doi.org/10.1093/hmg/ddu191 (2014). 45. Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 9, 559, https:// doi.org/10.1186/1471-2105-9-559 (2008). 46. Langfelder, P. & Horvath, S. Eigengene networks for studying the relationships between co-expression modules. BMC Syst. Biol. 1, 54, https://doi.org/10.1186/1752-0509-1-54 (2007). 47. Ritchie, S. C. et al. A scalable permutation approach reveals replication and preservation patterns of network modules in large datasets. Cell. Syst. 3, 71–82, https://doi.org/10.1016/j.cels.2016.06.012 (2016). 48. Reimand, J. et al. g:Profler-a web server for functional interpretation of gene lists (2016 update). Nucleic. Acids Res. 44(Web Server issue), W83–89, https://doi.org/10.1093/nar/gkw199 (2016). 49. Zhang, B. & Horvath, S. A general framework for weighted gene co-expression network analysis. Stat. Appl. Genet. Mol. Biol. 4, Article17, https://doi.org/10.2202/1544-6115.1128 (2015). 50. Zhang, K. et al. ICSNPathway: identify candidate causal SNPs and pathways from genome-wide association study by one analytical framework. Nucleic. Acids Res. 39(Web Server issue), W437–443, https://doi.org/10.1093/nar/gkr391 (2011). Acknowledgements Tis work was supported by the National Institutes of Health, US (R01 NR013391, R01 HL127332, and U01 HL065899) to KGT, by the Seoul National University Hospital [0420153010 (2015–1079)] to PHW, and by the Kangdong Sacred Heart Hospital Fund (2015-10) to LSY. Author contributions Heung-Woo Park, Suh-Young Lee, and Kelan G. Tantisira: Conception and design of the study, data generation, analysis, interpretation, and preparation of the manuscript. Ha-Kyeong Won, Byung-Keun Kim, Sae-Hoon Kim, Yoon-Seok Chang, and H. William Kelly: Data generation, and critical revision of the manuscript. Sang-Heon Cho: Data generation, analysis, interpretation, and critical revision of the manuscript. Competing interests Te authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41598-019-56656-9. Correspondence and requests for materials should be addressed to H.-W.P. Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Cre- ative Commons license, and indicate if changes were made. Te images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Scientific Reports | (2019)9:20133 | https://doi.org/10.1038/s41598-019-56656-9 10